Notes on morphology and morphophonemic analysis
1
Morphemes and phonemes
Morphemes are not the same as phonemes. A phoneme is the smallest unit that
distinguishes meaning, a morpheme is the smallest unit that has a meaning.
betting
bedding
Allophones [bERIN]
[bERIN]
Phonemes
/bEtIN/
/bEdIN/
Morphemes /bEt/ + /IN/ /bEd/ + /IN/
2
Notation
Morphemes are composed of phonemes, hence they are often expressed between
slashes /bEd/. The boundary between morphemes is often represented with plus
sign +, although a hyphen - is also used.
cat
betting
unidentifiability
/kæt/
/kæt/
/bEt + IN/ /2n + aIdEnt + Ifai + 2b@l + Iti/
/bEt-IN/
/2n-aIdEnt-Ifai-2b@l-Iti/
Morpheme boundaries sometimes coincide with word boundaries, but not always.
Word boundaries are often indicated with a hash sign #.
cat
betting
unidentifiability
/#kæt#/
/#bEt + IN#/
/#2n + aIdEnt + Ifai + 2b@l + Iti#/
3
Types of morphemes
The basic distinction between morphemes is those which are roots versus everything
else. The root is the central ‘core’ of the word. For a word like cat, the root is cat.
For a word like catty, the root is also cat. For a word like concatenation, the root is
concatenate.
Morphemes which attach to or otherwise modify the root are called affixes. Affixes can come before, after, or inbetween other morphemes, in which case they may
be referred to as prefixes, suffixes and infixes, respectively. The -ty in catty and
the -tion in concatenation are both suffixes; the un- in unidentifiability is a prefix.
Infixes are affixes that are inserted into a root, such as nominalising -(A)m in Khmer:
Verb/Adj
[kc@i]
[kh EN]
[Paoj]
4
N
‘unripe’
‘angry’
‘give’
[kamc@i]
[kamhEN]
[Pamnaoj]
‘something unripe’
‘anger’
‘gift’
Morphological functions
Morphology can largely be classified as either derivational (word-building) or inflectional (grammatical). Derivational morphemes are things like -able, -tion or -ity
in English, -keit in German, -ov(ý) in Czech, etc.
Inflectional morphology modifies a root to make it grammatically appropriate for
the context. Tense, aspect, and mood of verbs, number and person agreement in
nouns, and case in pronouns are examples of inflectional morphology.
Many languages, such as Mandarin, lack any kind of overt inflectional morphology;
the verb 给 [gě] ‘go’ in 我给了他一本书 ‘I gave him a book’ is the same as in 他给了我
一本书 ‘He gave me a book’. English has slightly more inflectional morphology, but it
is still a relatively impoverished system compared to many other languages. Compare
for instance the various English and Czech forms of the verb ‘to read’ (Czech forms
given in standard orthography):
2
English
1sg Pres
2sg Pres
3sg Pres
1pl Pres
2pl Pres
Czech
I read
čtu
you read
čteš
he/she it reads čte
we read
čteme
you (all) read
čtete
English
3pl Pres
Inf
3sg Imp
2pl Imp
1pl Imp
Czech
we read
čteme
to read
čı́st
read!
čti!
read!
čtěte!
let’s read! čtěme!
In Czech, the form of the verb changes in each case, overtly encoding information
about person and number.
5
Morphological analysis
The first step in a morphophonological analysis is often to figure out what the morphemes are. The basic approach is to examine the data and look for sequences of
phonemes which remain the same when two elements of meaning remain the same.
Consider the following data from Swahili:
atanipenda
atakupenda
atampenda
atatupenda
atawapenda
nitakupenda
nitampenda
utanipenda
utampenda
‘he will like me’
‘he will like you’
‘he will like him’
‘he will like us’
‘he will like them’
‘I will like you’
‘I will like him’
‘you will like me’
‘you will like him’
atanipiga
atakupiga
atampiga
ananipiga
anakupiga
anampiga
amenipiga
amekupiga
amempiga
‘he
‘he
‘he
‘he
‘he
‘he
‘he
‘he
‘he
will beat me’
will beat you’
will beat him’
is beating me’
is beating you’
is beating him’
has beaten me’
has beaten you’
has beaten him’
Table 1: Swahili verbs.
The first column are all forms of the verb ‘like’, the second column all forms of
the verb ‘beat’. Since the form penda is shared by all the words in the first column
and piga by all the forms in the second, we can probably conclude that penda means
‘like’ and piga means ‘beat’.
The next step is to isolate morphological minimal pairs, which are much like
phonological minimal pairs, except instead of finding pairs that differ in only one
3
phoneme, we are looking for forms that differ in terms of only one morpheme. For
example, the first two forms in the data atanipenda ‘he will like me’ and atakupenda
‘he will like you’ form a minimal pair allowing us to identify the 3sg-object and 2sgobject morphemes:
atanipenda
atakupenda
The remaining morphemes (for 3sg subject, 2sg subject, 1sg object, etc.) can be
discovered in a similar fashion.
6
Morphophonemic analysis: Finnish
Morphophonemic analysis proceed from the assumption (or the desire to assume) that
morphemes are stored in the lexicon in invariant, underlying forms (URs), which are
then converted into surface forms (SRs) by the application of phonological rules (or
constraints, in some theories). The analytic procedure is similar to that of phonemic
analysis, except that instead of only consider the distribution of phones, morphophonemic analysis also considers the distribution of morphemes.
Procedure for morphophonemic analysis
1. Make a provisional segmentation of words into morphemes
2. Find the allomorphs of each alternating morpheme
3. Within each allomorph, find the segment(s) that alternate
4. Set up the UR so that all allomorphs can be derived from a single UR by general
phonological rule
We’ll illustrate these steps using some data on Finnish nouns and adjectives (Table
2). We’ll compare two forms of each verb, the nominative case (used to mark the
subject) and the partitive (used to mark unknown amounts and irresultatives). The
goal of this analysis is to account for a alternation that gives rise to different vowels
in the pronunciation of two related forms.
4
Nom sg
Part sg
gloss
Nom sg
Part sg
gloss
aamu
hopea
katto
kello
kirya
kylmæ
koulu
lintu
hylly
kømpelø
kaikki
aamua
hopeaa
kattoa
kelloa
kiryaa
kylmææ
koulua
lintua
hyllyæ
kømpeløæ
kaikkea
‘morning’
‘silver’
‘roof’
‘clock’
‘book’
‘cold’
‘school’
‘bird’
‘shelf’
‘clumsy’
‘all’
yoki
kivi
muuri
naapuri
ovi
tukki
æiti
yærvi
koski
reki
væki
yokea
kiveæ
muuria
naapuria
ovea
tukkia
æitiæ
yærveæ
koskea
rekeæ
vækeæ
‘river’
‘stone’
‘wall’
‘neighbor’
‘door’
‘log’
‘mother’
‘lake’
‘waterfall’
‘sledge’
‘people’
Table 2: Finnish verbs.
A bit of background information. Although there aren’t any minimal pairs in the
data above, there is no evidence on which to conclude that any of the vowels are
allophones of any of the others. We will thus assume that Finnish has 8 phonemic
vowels /i y o ø æ u a e/. The first step is to try and determine what the relevant
morphemes are. In this case, we’re interested in the realisation of the nominative and
partitive singular morphemes.
If we consider the first 6 pairs in the first column, we can see that the difference
appears to be the presence versus absence of a suffix -a in the partitive. This suggests
that that partitive affix is -a and the nominative ‘suffix’ is -∅, or in other words, the
UR of the verb stem is the same as the nominative. This hypothesis is challenged
when we examine pairs like [kylmæ] ∼ [kylmææ] ‘shelf’ or [kømeplø] ∼ [kømpeløæ]
‘clumsy’. However, it turns out that this alternation is predictable based on other
properties of the stem, so it sometimes surfaces as -a and sometimes surfaces as -æ.
For the purposes of this analysis, this factor is not relevant, so we will designate the
underlying representation of the partitive as /-A/. Remember, this is an abstract
representation – there is no surface allophone [A], but that doesn’t matter. We just
want a way to represent that these two alternants are related by rule.
However, there is another alternation that emerges between the stems that can be
seen in pairs like [kaikki] ∼ [kaikkea] ‘all’, [ovi] ∼ [ovea] ‘door’, and [kivi] ∼ [kiveæ]
‘stone’: in some pairs, the final vowel of the stem alternates between -i and -e . That
is to say: all of the pairs in the data set fall into one of two groups: one where the
5
nominative form ends in -i and the partitive form, minus the suffix (-a or -æ) ends
in -i as well, and one where the nominative form ends in −i but the partitive form,
minus the suffix, ends in -e. To account for this alternation, we want to posit a single
underlying form for each verb stem and a single phonological rule that, when
applied to each verb form, generates the correct output.
Just on the basis of this observation, we could start to fomulate a hypothesis about
the underlying representation of the stems. One possibility would be to suppose that
the underlying forms of the stems are the same as the nominative forms, and there
is a rule which changes i to e in some environment – say, when followed by a [-back]
vowel. Let’s call this the i-lowering rule.
i → e/
V [-back]
(1)
We can check whether this rule works by applying it to a number of putative
underlying forms.
UR
i-lowering
SR
/#koulu#/
–
[koulu]
/#muuri#/
–
[muuri]
/#ovi#/
–
[ovi]
UR
i-lowering
SR
/#koulu+a#/
–
[koulua]
/#muuri+a#/
muure + a
*[muurea]
/#ovi+a#/
ove + a
[ovea]
Table 3: Derivations of surface representations using rule (1).
The first row shows three nominative forms and the SRs that result from the
application of the i-lowering rule (1). Since the structural description of the rule is
not met in any case, the rule does not apply, and the SR is the same as the UR. So far,
so good. However, when we apply the rule to the partitive forms in the second row, we
correctly generate forms like ovea, but incorrectly generate forms like muurea (which
should be muuria, with no vowel change). So the i-lowering rule, as formulated in
(1), is insufficient.
We could go on like this, positing rules and checking them against all the forms
in the data set (or, perhaps, write a short computer program to perform this tedious
task on our behalf). However, this is a somewhat haphazard approach that could
6
potentially take a great deal of time. In this case, we can quickly ‘zero in’ on the
relevant rule by observing some general properties of the data set. Recall that we
were told Finnish has eight vowels /i y o ø æ u a e/. Looking at the data, we find
that all of these vowels occur at the ends of words except [e]. Since the alternation
in question is between e and i, and e never appears word-finally, we can formulate a
rule that reflects this observation. We’ll call this the e-raising rule.
e → i/
#
(2)
If we assume that the UR of the stem has the vowel that surfaces in the partitive
forms, we can use rule (2) to correctly derive all the forms in the data.
UR
e-raising
SR
/#koulu#/
–
[koulu]
/#muuri#/
–
[muuri]
/#ove#/
ovi
[ovi]
UR
e-raising
SR
/#koulu+a#/
–
[koulua]
/#muuri+a#/
–
[muuria]
/#ove+a#/
–
[ovea]
Table 4: Derivations of surface representations using rule (2).
On this formulation, the UR is generally reflective of the surface form; the rule is
necessary just to correctly derive the nominative forms of verbs that have an e-final
stem in the partitive. Like the Polish devoicing rule, this rule is neutralising since
the phonemic distinction between /e/ and /i/ is neutralised in word-final position:
an underlying /e/ becomes a surface [i].
Note also that this provides another example of why it can be dangerous or at
least misleading to assume that the ‘most basic’ or ‘simplest’ surface form of a word
based on morphological criteria (such as being the nominative case or the 1st person
singular) is straightforwardly related to the underlying representation.
7
7
Morphophonemic analysis: Kerewe
Consider the following verbal data from Kerewe, a Niger-Congo (Bantoid) language
spoken in Tanzania.
infinitive
1sg habitual
3sg habitual
imperative
kupaamba
kupaaNga
kupima
kupuupa
kupekeča
kupiinda
kuhiiga
kuheeka
kuhaaNga
kuheeba
kuhiima
kuhuuha
mpaamba
mpaaNga
mpima
mpuupa
mpekeča
mpiinda
mpiiga
mpeeka
mpaaNga
mpeeba
mpiima
mpuuha
apaamba
apaaNga
apima
apuupa
apekeča
apiinda
ahiiga
aheeka
ahaaNga
aheeba
ahiima
ahuuha
paamba
paaNga
pima
puupa
pekeča
piinda
hiiga
heeka
haaNga
heeba
hiima
huuha
‘adorn’
‘line up’
‘measure’
‘be light’
‘make fire with a stick’
‘be bent’
‘hunt’
‘carry’
‘create’
‘guide’
‘gasp’
‘breathe into’
Table 5: Kerewe verbs.
Posit an underlying representation for each verb stem and formulate a single rule
that correctly derives all of the surface forms.
8